Vesall Nourani is a PhD Candidate in Economics at Cornell University and a National Science Foundation Graduate Research Fellow; he is on the job market this year
It’s the mid 1990’s in Ghana and pineapple farmers are learning how to properly apply fertilizer by seeking advice from their friends. It is obvious to many that the high profit margins from pineapple farming are making residents in some villages wealthier; however, in other villages this relationship is not so clear – despite the rising demand for pineapple from export markets, some pineapple farmers have disadopted (stopped farming) the crop. Their decision is noticed by all village residents. After all, disadopters have expended much time and energy to first plant and then remove a perennial sucker plant from their fields. Something must not have gone as planned! A farmer who was once considering planting pineapple himself takes others’ disadoption as a warning and no longer considers pineapple worthy of his attention.
Contribution#1: Farmers Learn from Disadopters of a New Crop
There is a large literature that examines whether farmers are influenced to adopt new technologies from peer experiences (Foster and Rosenzweig, 1995; Munshi, 2004; Bandiera and Rasul, 2006; Conley and Udry, 2010). Nevertheless, the main narrative in this literature does not account for the possibility of disadoption and, consequently, the possibility that one learns from disadopters. The main narrative goes something like this: a new crop variety is assumed to be unambiguously more profitable than traditional varieties, but lack of production knowledge prevents its widespread adoption. When opportunities for learning start to emerge – e.g., someone in a social network adopts a new crop – farmers gain production knowledge, which allows them to benefit from the new variety.
As someone who (for better or worse) resists the latest fads, I find the assumption of superiority of new technologies somewhat troubling. Indeed, there are well-founded reasons to feel this way: new technologies do not always provide positive returns to farmers (see Duflo et. al (2008), Marenya and Barrett(2009), Suri (2011) and Magnan et. al (2015) among others). Therefore, rational farmers should have no reason to believe in the superiority of a new variety prior to adopting. A farmer who adopts a new crop variety and subsequently experiences consistently lower profits than expected will likely stop using (disadopt) the new variety. How do we introduce the social influence of disadopters into a model of social learning?
Social Learning: “Whether to Adopt” or “How to Farm” a New Crop?
The modelling approach I’ve taken in my paper is to distinguish between learning “whether to” adopt a new crop from “how to” farm the new crop. Before learning how to farm a new crop, farmers determine whether they believe the new crop is more profitable than other options. Changing one’s beliefs regarding the viability of a new variety is cheap relative to learning how to farm the new crop (for example, Hanna et. al (2014) show how difficult it is to notice aspects of a production function that can substantially increase seaweed farming profits). Production knowledge requires costly learning effort. The decision to seek out production knowledge depends in large part on how costly it is to acquire, but also on whether one believes such a knowledge investment is worth the cognitive effort in the first place! This framework allows me to conceive of two stages of learning: first, one determines whether the new variety is relatively profitable. If it is, one proceeds to learning how to produce the new variety.
Contribution #2: Learning from friends and reacting to acquaintances
I return to the site of a famous study of pineapple farmers in Ghana by Chris Udry and Tim Conley (2010) who demonstrated that farmers gain production knowledge from one another’s farming experiences. In 2009, Chris Barrett (Cornell University) and Tom Walker (World Bank) collected data on the history of pineapple adoption and disadoption by the same households as the above study. Additionally, they collected data on a within-sample census of network links, differentiating between close friends and more distant acquaintances, in each of the four villages, allowing me to analyze whether farmers with more disadopting peers had a lower probability of adopting the new crop.
The ability to distinguish relationship type in the data is a key means of distinguishing between the two objects of social learning in my model. Given the cost of acquiring production knowledge, farmers are more likely to seek out such knowledge from those they are comfortable interacting with regularly, such as a close friend. Farmers will only seek this knowledge, however, when they believe the new variety to be relatively more profitable than status quo profits. When it comes to belief dynamics, however, distant friends play a crucial role. Because close friends communicate beliefs on a year-to-year basis, their observable adoption or disadoption decisions are somewhat expected. As a result, the moment of adoption or disadoption does little to influence profitability beliefs. Distant acquaintances, on the other hand, are less likely to communicate regularly with one another. Thus, their observable adoption or disadoption decisions signal that their beliefs have changed over the years, representing information that rational farmers will use to update their own beliefs.
The above figure estimates a hazard model that is closely linked to my theoretical model and shows interesting divergence in the estimated social influence of adopters and disadopters according to relationship type: distant friends (DF) and close friends (CF). The left-hand panel describes how adoption probability changes as a function of the previous year’s change in the average share of adopters or disadopters in a “distant friend network” (a similar figure depicting “close friend networks” yields null results as hypothesized). In my paper, I describe how the average share of adopters/disadopters is more likely to influence belief dynamics than opportunities to acquire production knowledge – the latter is proxied by the total number of peers with past or current experience in pineapple farming (right-hand panel).
In the left-hand panel, disadoption by distant friends decreases a given farmer’s adoption probability in any given year – a 10 percentage point increase (from zero percent) in the share of network disadopters is associated with a nearly 50% decrease in adoption probability. This effect is larger in magnitude than adoption by distant friends, suggesting that beliefs change more following network disadoption than adoption. The right-hand panel demonstrates that an increase in close friends with pineapple farming experience is associated with a 6% increase in adoption probability on average – this effect is not statistically significantly present among distant friends. This latter effect persists if one removes the number of adopters and only considers the influence disadopting close friends and is only present during years in which pineapple farming was likely to be profitable in Ghana (indicated by a negative shock to the prices of MD2 pineapple in 2004).
What do we gain from delineating two objects of social learning in this manner? First, cash-strapped extension delivery services are always looking for a bigger bang for their buck. As a result, it is quite useful to understand when and how social networks magnify the influence of an extension agent (e.g., Beaman et. al, 2015). However, the possibility of disadoption might induce us to ask whether social multipliers might work in the opposite direction – leading communities to a path-dependent state in which no one adopts a new variety because an early attempt failed. Much more empirical investigation is needed to determine the extent to which such patterns takes place.
Alternatively, we might consider the relationship between the innovation of new technologies and the capacity of communities to learn about new technologies. Griliches (1957) links the technology adoption problem with the technology innovation problem. Innovations require experimentation, and the diffusion of the practice of experimentation are likely to be negatively correlated with learning costs. Researchers and practitioners might look into whether locally appropriate agricultural innovation, through the creation of local experimentation and research networks, may decrease learning costs enough to encourage more farmers to acquire production knowledge; the type of production knowledge that likely leads to greater productivity and a path out of poverty.
*This post first appeared at Development Impact on December 13, 2017 as a part of their 2017 Blog Your Job Market Paper series.